节点文献

凡纳滨对虾集约化养殖水质管理决策支持系统的构建

The Construction of DSS in Water Quality Management of Intensive Litopenaeus Vannamei Shrimp Tanks

【作者】 马真

【导师】 万荣; 宋协法;

【作者基本信息】 中国海洋大学 , 增殖养殖工程, 2013, 博士

【摘要】 集约化养殖带来巨大经济效益的同时,也给自身的可持续发展和生态环境带来了严峻的挑战。对养殖企业而言,关键水质因子快速监测和对水质异常情况的及时识别,是养殖安全的重要保障。目前养殖水质管理的主要内容是对水质监测数据进行整理和汇总,然后分析水质的变化情况。但目前凡纳滨对虾养殖的水质管理大多是针对养殖水质的当前或历史数据进行的,得到的结果只能反映水质的历史状况,水质管理处于一种滞后和被动的局面,往往造成严重后果时才发现水质状况恶化严重,所以必须改变传统的管理方式,在对水质进行实时监测、现状评价的基础上对水质进行预测,最终对其恶化趋势进行预警,并制定相应的应急措施减缓或阻止水质恶化,实现真正意义上水质管理。本研究以凡纳滨对虾集约化养殖生产实际为基础,在分析了凡纳滨对虾集约化养殖水质特征基础上,首先建立了凡纳滨对虾集约化养殖水质评价模型,克服了以往评价方法的不足,能够对养殖水质做出综合评价。然后以人工神经网络(ANN)为手段,建立了凡纳滨对虾集约化养殖水质预测模型,实现了养殖关键水质因子与预测对象之间的映射,解决了养殖水体变化过程中存在的非线性、多变量和模糊性问题,提高了水质预测的精度。第三,建立了凡纳滨对虾集约化养殖水质快速预警模型,将单因子预警与多因子预警相结合,对集约化养殖过程中潜在的危险进行快速准确预警。第四,完成凡纳滨对虾集约化养殖水质管理决策支持系统,实现水质科学管理的可视化,能够为养殖管理者进行水质调控和实现科学管理提供有效的技术手段。本文的主要研究结果如下:1.凡纳滨对虾集约化养殖水质特征本实验主要研究了凡纳滨对虾集约化养殖水质相关的各水质因子的特征。实验共监测了11项水质指标,在对实测数据进行显著性检验和统计处理后,确定了影响对虾集约化养殖水质的主要因子,然后采用多元逐步回归分析了凡纳滨对虾集约化养殖主要水质因子与其它水质因子间的相关关系,并用回归方程的形式予以表示。结果表明,多元回归模型能够给出某水质因子受其它水质因子影响的定量公式。此外,与一般的多元线性回归模型相比,多元逐步回归模型能够分辨出与需描述变量显著相关的变量及其干扰变量,只将对水质因子影响最大的因子列入回归模型,提高了模型的可靠性。2.凡纳滨对虾集约化养殖水质评价模型本部分主要介绍了因子分析模型建立的一般步骤以及凡纳滨对虾集约化养殖水质评价模型的建立及应用。以水质实测数据为基础,结合我国《渔业水质标准》和国内外各相关养殖水质标准,确定了凡纳滨对虾集约化养殖水环境质量分类标准。以因子分析为基础,建立了凡纳滨对虾集约化养殖水质评价模型,并对模型进行了应用,模型表达式为:zF=W1*F1-W2*F2-¨¨-Wn*Fn。结果显示,本模型不仅能够体现水质因子之间的相互关系,客观反映水质的真实状况,还能够在某一指标特别恶化时对水质做出准确评价,弥补了以往评价方法必须结合单因子评价方法的不足。3.凡纳滨对虾集约化养殖水质预测模型本部分主要介绍了BP神经网络的原理、建立模型的一般步骤以及凡纳滨对虾集约化养殖水质预测模型的建立及仿真过程。将第二章确定的九项水质因子作为凡纳滨对虾集约化养殖水质BP预测的输入向量;根据预测目的,确定BP预测网络的输出变量为水质预测值;网络采用三层结构,隐含层的神经元数为7;tansig和purelin分别作为网络隐含层和输出层神经元的传递函数。在确定了凡纳滨对虾集约化养殖水质BP预测模型的结构后,对模型进行训练,选择trainlm作为网络的训练函数,最后通过仿真实验对模型进行验证。结果显示,两组数据相关分析显示系数是0.9921,预测误差率结果显示平均的预测误差率是2.9%,最大的是12.55%,最小是0.038%,总体预测结果较好。BP神经网络能够以较高的精度对养殖水质状况进行预测,使得集约化养殖水质预测成为可能,能够实现水质恶化的早期预报,减少养殖损失,保证养殖的安全。4.凡纳滨对虾集约化养殖水质预警模型根据对虾养殖水质的特征,建立了凡纳滨对虾集约化养殖水质单因子预警模型和水质综合预警模型。在确定了预警指标体系和预警级别的基础上,以凡纳滨对虾养殖水环境质量分类标准为基础,建立了水质单因子预警模型,并在传统的单因子预警模型基础上,以第二章建立的多元逐步回归模型为手段,建立了对虾养殖水质单因子预警回归模型,其不仅继承了传统单因子预警模型简单直观的特点,而且还能客观的能够反映出某水质因子与其它水质因子之间的关系。建立了凡纳滨对虾集约化养殖水质综合预警模型,并将状态预警与趋势预警相结合,在水质评价模型基础上建立了状态预警模型,模型的表达式为::E=W1×F1-W2×F2-¨¨-Wn×Fn。在预测模型基础上建立了趋势预警模型,BP神经网络预警模型的结构为9-7-1,网络隐含层和输出层神经元的传递函数分别为tansig和purelin, trainlm为网络的训练函数,最后通过仿真实验对模型进行验证,趋势预警结果显示,预警结果与实际结果数据相关分析系数为0.991,说明总体预警效果较好。5.凡纳滨对虾集约化养殖水质决策支持系统在上述模型建立的基础上,设计并实现了凡纳滨对虾集约化养殖水质决策支持系统。系统使用的是基于模型的辅助决策系统,在确定了框架和运行结构后,结合实验实际情况,通过随机抽取的水质数据,验证DSS在用于水质评价和预警这两个模块中的应用。

【Abstract】 The intensification of aquaculture has brought substantial economic benefits, aswell as promoted waste production and disease outbreak. In recent years, deterioratingwater quality has caused massive financial losses to farmers, and has become one ofthe major bottlenecks to production output. For aquaculture enterprises, consideringthe innumerable and complicated variations in water quality, monitoring programsand the reliable estimation of water quality play important roles in culturemanagement to provide a thorough understanding of the degree of contamination andto limit its effect. The data were collected and analyzed, which was the presentcontext of water quality management. Traditional approaches for water qualitymanagement do not provide a comprehensive view of overall water quality. By thetime we’d scoped out the problem, it was too late. It requests that the water qualitymanagement must change the pattern of the traditional management and make thealternative into the way of the real-time monitoring, current situation evaluation andin time alarming.In the present study, the water quality of intensive culture tanks of L. vannameiwas analyzed. Firstly, on the basis of analyzing the weekly values of water qualityvariables measured in the shrimp farm, the intensive culture water quality assessmentmodel was established. It was a practical tool for fast and easy data interpretation, andits application in monitoring the quality of the water sources is recommended for themanagement of shrimp farming or other production activities. Secondly, The ANN model is built for forecasting of shrimp water quality in intensive culture tanks. Themodel can describe complex nonlinear effects between water quality variables andwater quality. Thirdly, the early warning model was established. The present studycombined single-factor warning model with multi-factor warning models. Finally, thedecision support system was achieved. The main results are as follows:1. The water characteristics of the intensive shrimp cultureThe present study investigated the characteristics of water quality parametersrelated to shrimp water quality. Eleven different water quality parameters wereanalyzed during the experimental period. A stepwise multiple regression model wasused to identify the significant correlation among water quality parameters. The resultshowed that the correlation between the water quality parameter and the otherparameters was studied quantitatively the quantitative formula was fitted. In addition,compared with the multiple linear regression model, stepwise multiple regressionmodel could identify the main variables and interference variables, improved thecredibility and reliability of model.2. Water quality assessment model for intensive shrimp tanksIn the present study, the water quality of intensive culture tanks of L. vannameiwas evaluated using factor analysis model. According to the weekly values of waterquality variables measured in the shrimp farm and fisheries water quality standards athome and abroad, the water quality criteria of Litopenaeus vannamei for intensiveshrimp tanks were determined. The source identification indicated that the variablesresponsible for water quality deterioration in the intensive culture shrimp tanks weremainly related to organic matter, natural condition, and nutrient. The nine waterquality variables remained were chosen and the final equation waszF=W1*F1-W2*F2-¨¨-Wn*Fn.In summary, the model was a practical toolfor fast and easy data interpretation, and its application in monitoring the quality of the water sources is recommended for the management of shrimp farming or otherproduction activities.3. Water quality forecasting model for intensive shrimp tanksWe used a backpropagation neural network (BP-NN) model to predict the waterquality in intensive Litopenaeus vannamei shrimp tanks. It was developed usingmeasured water quality data that were generated over120days with weeklymonitoring in four different shrimp tanks. Nine parameters were selected as inputvariables: water temperature, pH, total ammonia nitrogen, nitrite nitrogen, nitratenitrogen, dissolved inorganic phosphorus, chlorophyll-a, chemical oxygen demand,and five-day biochemical oxygen demand. The model has a tan-sigmoid transferfunction for the hidden layer and a linear transfer function for the output layer. TheLevenberg–Marquardt algorithm was used to overcome the shortcomings of thetraditional BP algorithm; that is, low computational power and getting stuck in localminima. The number of hidden layer nodes was optimized by a trial and errorapproach, and seven optimal neuron nodes were identified. The computed results forwater quality show good agreement with the experimental values. The correlationcoefficient of the data set is0.9921. The simulation results reveal that the BP-NNmodel efficiently predicts the water quality in intensive shrimp tanks.4. Water quality early warning model for intensive shrimp tanksThe single-factor and multi-factor early warning model were established on thebasis of the characteristics of water quality parameters. After the level of warning hasbeen identified, the single-factor early warning model was established. The modelbased on the water quality criteria of Litopenaeus vannamei for intensive shrimp tanks.In addition, we were modeling the single-factor early warning model based on thestepwise multiple regression. The multi-factor early warning model consisted of twoparts: status early-warning and trending early-warning. The status early-warning established on the basis of the water quality assessment model. The final equationwas=W1×F1-W2×F2-¨¨-Wn×Fn. The trending early-warning established onthe basis of the water quality forecasting model. All available variables were selectedas input variables. The model has a tan-sigmoid transfer function for the hidden layerand a linear transfer function for the output layer. The Levenberg-Marquardtalgorithm was used to overcome the limitations of the traditional BP algorithm. Thenumber of hidden layer nodes was optimized by trial and error. The computed resultsfor water quality were in good agreement with the experimental values. Thecorrelation coefficient (R2) of the data set was0.991. The simulation results revealthat the BP-NN model efficiently predicts the water quality in intensive shrimp tanks.5. The application of decision support system in water quality managementBased on the above models, the decision support system in water qualitymanagement was designd and established. The paper used supplement decisionsupport which based on model. After the framework and running structure weredetermined, we show several examples of how the DSS is used to water qualityassessment and alarming.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络